Risk-sensitive control as inference with Rényi divergence
This provides a theoretical foundation for risk-sensitive reinforcement learning, though it appears incremental as it builds on existing control as inference and maximum entropy control frameworks.
The paper tackles the problem of risk-sensitive control in reinforcement learning by introducing a framework that extends control as inference using Rényi divergence, showing it unifies risk-neutral methods and derives algorithms like policy gradient and soft actor-critic.
This paper introduces the risk-sensitive control as inference (RCaI) that extends CaI by using Rényi divergence variational inference. RCaI is shown to be equivalent to log-probability regularized risk-sensitive control, which is an extension of the maximum entropy (MaxEnt) control. We also prove that the risk-sensitive optimal policy can be obtained by solving a soft Bellman equation, which reveals several equivalences between RCaI, MaxEnt control, the optimal posterior for CaI, and linearly-solvable control. Moreover, based on RCaI, we derive the risk-sensitive reinforcement learning (RL) methods: the policy gradient and the soft actor-critic. As the risk-sensitivity parameter vanishes, we recover the risk-neutral CaI and RL, which means that RCaI is a unifying framework. Furthermore, we give another risk-sensitive generalization of the MaxEnt control using Rényi entropy regularization. We show that in both of our extensions, the optimal policies have the same structure even though the derivations are very different.